To address the high computational cost of CFD simulations for unsteady aerodynamic forces on rotorcraft trailing-edge flaps, a data-driven reduced-order model based on one-dimensional convolutional neural networks (1D-CNN) is developed. Taking the BO105 rotor NACA23012 airfoil with a trailing-edge flap as the research object, training data are obtained using CFD methods. The 1D-CNN is employed to learn the nonlinear mapping relationships between the time-series signals of the main airfoil pitching motion and the trailing-edge flap deflection motion and the lift coefficient, drag coefficient, and pitch moment coefficient, thereby constructing an efficient aerodynamic prediction model that can directly replace CFD. The feasibility of the model is verified through static cases, pressure distribution, and deep stall cases. Test results demonstrate that the proposed reduced-order model accurately captures strongly unsteady aerodynamic characteristics and significantly improves computational efficiency compared to traditional CFD simulations. Finally, aerodynamic predictions for wide pitch amplitudes in the deep dynamic stall regime are performed, and the model’s results agree well with those obtained from the CFD solver.